Randomised controlled trials (RCTs) in surgery are uncommon and often more complex and difficult to design and implement than trials of pharmaceutical interventions. The reasons for this include challenges with recruitment (due to lack of clinical equipoise or treatment preference), difficulties with standardisation and delivery of interventions between trial centres or individual surgeons (particularly if the interventions under evaluation are new) [1, 2], and challenges concerning the appropriate selection, measurement and reporting of outcomes.

An outcome (or endpoint) is a direct or indirect measurement of the effect of an intervention on a participant’s clinical or functional status [3, 4]. The purpose of outcome measurements in an RCT is to provide information about the effect of an intervention under evaluation compared to a standard procedure or control. In a trial, participants are randomised to receive different interventions with the aim of negating the effect of confounding factors. In this way, any observed differences in outcome between the treatment groups can be attributed to the effect of the intervention and not to a baseline characteristic of the participant or other potentially influential variable. When designing a clinical trial, it is necessary to define a trial hypothesis of the expected effect of the intervention on a primary outcome (the outcome of greatest interest). This allows the calculation of a sample size which is essential to inform the number of centres in the trial and the length of trial recruitment. In some trials, it may be advantageous to have a composite primary outcome comprised of more than one outcome of interest. This can be advantageous when event rates are low, to allow for smaller sample size calculation and increased statistical efficacy, or when the choice between several important outcomes is arbitrary [5]. Limitations of using a composite outcome, however, include the need for a more complex statistical analysis plan and ensuring that all component outcomes have equal importance to patients [6, 7]. Secondary outcomes are measured to evaluate the additional effects of the intervention and answer other relevant questions [8]. They may also be chosen for exploratory purposes in order to develop a hypothesis for future research. Outcomes may be classified with regard to their type (primary, secondary, or exploratory), and whether they are assessed objectively or influenced by patient or clinician judgement [9]. They may also be categorised by the domain of measurement. Clinical outcomes may be defined as outcomes measured in clinical practice such as survival/mortality or surgical complications and adverse events (e.g. anastomotic leak or deep vein thrombosis). Clinical outcomes can be completely objective (e.g. death) or they can include a professional assessment or observer interpretation (e.g. wound infection). Other types of outcomes may be patient-reported outcomes (PROs); assessments made directly by the patient themselves rather than an observer, such as symptom severity or reports of functional health status. Most PROs are recorded in self-completed questionnaires with validated rating scales that are standardised and scored and are often referred to as measures of health-related quality of life (HRQL). Whilst some outcomes may be referred to as HRQL outcomes, this term is confusing, and it is recommended that outcomes are described as PROs (measured by patients themselves), or observer assessed outcomes (measured by clinicians/researchers), rather than the ill-defined term HRQL—which may be measured by a patient or an observer. Surrogate outcomes are outcomes that predict another clinically meaningful endpoint of interest [10] and may be used because they are more practical or less invasive, for example, the measurement of coagulation factors or albumin as a surrogate marker of liver function in patients following liver surgery [11]. Other types of outcomes include hospital or ‘process’-related factors such as length of hospital stay or frequency of tests, and assessments of cost and resource use (Box 1).

Many of these different types of outcomes can be assessed once or on several occasions following surgery. The timing of outcome assessment varies in trials, and can be classified as during or after surgery, as short, medium or long-term. Short-term assessments are often made during the hospital stay and longer-term assessments maybe years after the intervention. In the selection of outcomes for RCTs in surgery, it is necessary to consider these different types of outcomes, who will assess them, and the appropriate time to assess them in light of the research question and trial design.

Historically, surgeons have selected and assessed outcomes that focus on short-term clinical events and those related to the surgical process such as complication rates and length of hospital stay [12]. This may be because these outcomes reflect aspects of patient outcome and recovery that are of importance to the surgeons designing the study, and reflect the quality of the delivery of the intervention and the surgeons’ technical skill [13]. They are also events that require early detection and treatment by surgeons, which may result in serious problems for patients if untreated. Whilst it is important to assess and report these short-term clinical outcomes, many surgical studies do not examine the longer-term benefits and harms of surgical interventions or consider outcomes from a patient perspective [14]. In addition, undue focus on operative complications may be unhelpful as these outcomes are often rare, and if trials are not powered sufficiently to exclude differences in uncommon complications, the data can be misleading. There is, therefore, a need for better standards of outcome selection, measurement and reporting in RCTs in surgery. This paper will consider some of the key challenges and suggest solutions to these problems, highlighting current initiatives working towards improvements in these areas.

Challenges for outcome measurement in RCTs in surgery

The methodological and practical challenges for the selection and measurement of outcomes in surgical trials include choosing the right outcome for the trial design and purpose, selecting relevant outcomes to measure from the range of possible outcomes and measuring them with minimal risk of bias.

Choosing relevant outcomes to suit the trial design and purpose

In order to select the most appropriate outcomes for any trial, it is first necessary to have a good understanding of the research hypothesis. This will determine the trial design and the appropriate outcomes to be measured. Trials can be broadly classified as explanatory or pragmatic in design. Explanatory (efficacy) trials are designed to test whether an intervention works under ideal conditions (performed by an experienced surgeon, for example, in a single centre with standardised pre- and post-operative care) and for these reasons, the intervention is delivered within a very restrictive protocol often on a highly select group of patients [2, 15]. Outcomes in explanatory trials are selected to provide information on whether the intervention works within these restricted conditions and tend to focus on measurable clinical or biological symptoms or markers [16]. For surgical trials, these are typically short-term clinical or process measures that supply data about the safety of the intervention and its immediate health benefits and risks. Examples include measured changes in intra-operative variables (e.g. length of operation, blood loss) or short-term adverse events and in-hospital morbidity (e.g. need for pain killers, length of hospital stay) (Table 1, [17]). Pragmatic clinical (effectiveness) trials, on the other hand, are designed to assess whether an intervention works in routine clinical practice and everyday settings. The intervention is delivered under more flexible conditions and by practitioners with a range of expertise to reflect typical practice. Members of the research team, such as the surgeons, often have more scope to make choices about the exact procedure performed, or the approach used (depending on the research intervention). The sample population is usually less restricted and in combination, these factors aim to make the results generalisable to a wide range of clinical environments. Pragmatic trials often provide information on the longer-term health gains and harms of an intervention and results are used to inform patients, clinicians and policy makers when making decisions on treatment. A wide spectrum of outcomes can be assessed during a pragmatic trial and an assessment of resource use may also be included. The majority of RCTs in surgery aim to determine if the intervention should be performed in standard clinical practice and therefore, a pragmatic trial design is often most appropriate [18]. However, reviews of the surgical literature show that the number of high-quality pragmatic trials is few and often focus on short-term and clinical measures providing uncertain value for patients, surgeons and other decision makers [1923]. Although there is a paucity of studies, there are examples of well designed and conducted pragmatic RCTs in surgery with appropriate outcome measures (Table 1). There are also examples of trials using composite outcomes combining short-term measures and including some longer-term assessments [24]. However, more high-quality RCTs in surgery are required with better trial design and outcome measures to allow the results to be relevant to routine clinical practice. This will require that surgeons involved in trial design to consider which outcomes are important to patients and include assessment of these in trials as well as the standard measures used to evaluate surgery.

Table 1 Short- and long-term outcomes commonly used in examples of RCTs in surgery

Multicentre, pragmatic trials require surgeons to work together and to involve patients and other health-care professionals in the trial design process at an early stage. The National Institute for Health Research (NIHR) in the UK funds an advisory group to support greater public and patient involvement in research [25] and the James Lind Alliance charitable organisation aims to bridge the gap between patients and researchers [26]. A similar initiative in Canada, the Patient-Centred Outcomes Research Institute, encourages the involvement of patients to shape the research agenda [27]. In the UK, recent funding opportunities from the NIHR and Royal College of Surgeons of England for Surgical Trial Centres [28] are also promoting collaborative working, and there are trainee research collaboratives whose purpose is to bring together surgical trainees and offer the opportunity to run multicentre trials [29, 30]. Drives to bring together surgeons and methodologists and to educate the surgical community in clinical trial design have been generated. Examples of this include the international IDEAL collaboration [31], the UK Medical Research Council Hubs for Trials Methodology Research [32] and the American College of Surgeons Continuous Quality Improvement Surgical Research Committee [33]. In Germany, the Study Centre for the German Surgical Society has similar aims to increase the number of well-designed multicentre surgical trials [34]. These initiatives should lead to improvements in the design of randomised trials of surgical interventions.

Selecting and reporting appropriate outcome measures for RCTs in surgery

After establishing the trial design and relevant outcomes, the next challenge for RCTs in surgery is to clearly define and select the outcomes to be measured from the range of possibilities. In surgical studies, the approach to outcome assessment has historically been assorted, inconsistent and ill-defined. A systematic review of adverse events after gastrointestinal surgery, for example, identified 56 different definitions and measures for anastomotic leak [35] and up to 10 different measures for mortality have been used in studies after oesophagectomy [23]. In colorectal cancer surgery, a systematic review found 766 different clinical outcomes were assessed, with inconsistency in selection, measurement and reporting [36]. This variation across trials and lack of standardisation is problematic for systematic reviews and meta-analyses of surgical interventions as outcomes are often not comparable, meaning synthesis and amalgamation of findings is seldom possible. In addition to inconsistent outcome measurements, other problems for RCTs in surgery are caused by the multiplicity of outcomes that are measured in a single trial. As a result, authors of trial reports do not always include the full range of outcomes that have been measured, often focusing on outcomes with ‘interesting’ or statistically significant results [37, 38]. Such selective reporting of outcomes leads to outcome reporting bias and causes further problems for systematic reviews by distorting the available evidence [3941]. Outcome reporting bias is a particular issue for PROs, where numerous multiple health domains are measured within a single questionnaire and those of particular interest are often not specified a priori or when PROs are secondary outcomes.

Core outcome sets (COSs) are one solution to the problems caused by multiple outcome selection and outcome reporting bias. Core outcome sets are a collection of the important (core) outcomes to be measured and reported in all pragmatic trials of a specific disease or condition, and their use and application allows for the results from multiple trials to be readily combined and compared. Core outcomes are agreed by consensus between key stakeholders such as patients, health-care professionals, and members of funding bodies. The Core Outcome Measures for Effectiveness Trials (COMET) Initiative in the UK supports the development and application of COSs and holds a database of all existing work in this area [42]. Several COSs to use in trials of surgical intervention are being developed, for example in oesophageal, colorectal and head and neck cancer as well as other diseases outside oncology [4244].

Assessing and measuring outcomes with minimal risk of bias

Once the outcomes to be measured are agreed upon and defined, it is necessary to ensure that they are assessed appropriately to reduce the risk of bias. Bias can exaggerate treatment effect, thereby making trial results unreliable and problematic when implementing for clinical practice [9]. Consideration must be given to how outcomes are measured and by whom. It is also important that the tools used to assess these outcomes are developed, tested, and validated properly, as the use of inappropriate tools result in unreliable or invalid data. This applies to any outcome measure being used in a trial, whether it be a clinical measure of anastomotic leak, for example, or a PRO.

There are several forms of bias from which surgical trials are particularly at risk, including performance bias and ascertainment bias. Performance bias results from differences in the way patients in the two treatment arms are managed post-operatively. It can have an effect on all types of outcomes, and is a particular problem in trials of surgical interventions because of the intimate role played in both delivery of the intervention and post-operative care provision by the operating surgeon. Ascertainment bias is a term used to describe collectively observer bias (when outcomes are reported by an outcome assessor) and reporting bias (for PROs), and can be a particular problem in trials whose primary outcome is a PRO. Both performance and observer bias can occur on a conscious or sub-conscious level and might result from pre-existing expectations about the efficacy of the treatments under evaluation, a lack of clinical equipoise or the influence of knowledge about treatment allocation. One method that can be used to minimise the risk of both performance and ascertainment bias is blinding (or masking). Blinding is used in RCTs to hide treatment allocation from various members of the research team or trial participants. In the ideal setting, blinding is maintained until all primary outcome data are collected. The decision about who should be blinded in a particular trial depends on the trial design, interventions and available funding. In pharmacological trials, it is common to use a matched placebo to blind staff and patients. In the case of surgical trials, however, using a placebo or sham surgery is often difficult or impossible because it may be considered unethical to subject patients to general anaesthesia and a mock operation for solely research purposes, although such examples exist [4551]. The complex nature of surgery can also make blinding of staff difficult because of the necessity for input from a large multidisciplinary team and it is frequently impossible for surgeons to be blinded because of their role in delivery of the intervention. In this case, where ever possible, it is recommended that the surgeon does not collect the outcome data and that it is collected by someone blinded to the intervention type.

Despite these challenges for surgical RCTs, novel methods may be effectively employed to blind some of the research team. Examples where this has been successfully achieved are illustrated in Table 2 [49, 5257]. In a trial whose primary outcome measure is a PRO, it is important to consider whether it is possible to blind the patient from the treatment allocation, although this may also be impossible. It may be more practical in this situation to blind outcome assessors who are ideally independent from the central research team. For example, an adjudicating committee may be used to review and assess digitalized clinical, photographic or imaging data with no details of the patient or intervention available to them [57].

Table 2 Methods to blind patients, surgeons and outcome assessors in examples of RCTs in surgery

Summary and conclusion

Well designed and conducted pragmatic RCTs are essential for the evaluation of surgical interventions. Central to this is the need to select, define, measure and report outcomes that are relevant to the trial design, valid and reliable, and to assess these outcomes with minimal ascertainment and performance bias. This can be achieved with surgeons and methodologists working closely together. The need to improve and increase the number of surgical trials has been internationally recognised and a cultural shift in the approach to trial design and participation is taking place. The development and application of core outcome sets for pragmatic RCTs in surgery will reduce many of the problems associated with the inconsistency and multiplicity of outcomes that have historically been assessed in RCTs in surgery and this is being addressed by work of the COMET initiative. Finally, the challenge to reduce and eliminate bias in surgical trials is being tackled by innovative methods to perform blinding to ensure that outcomes are fairly assessed and measured and effects of the intervention are not over or under estimated. These combined approaches and efforts will ensure surgical interventions are appropriately evaluated with informative outcomes, so that results can be useful for clinical practice.

Box 1. Types of outcomes and outcome measures used in RCTs in surgery

Clinical outcome measure

An outcome measured in clinical practice, e.g. survival or surgical complications such as anastomotic leak, including the specific measurement variable (e.g. systolic blood pressure), analysis metric (e.g. change from baseline, final value, time to event), method of aggregation (e.g. median, proportion) and timepoint [58].

Patient reported outcome (PRO)

An outcome reported directly by the patient themselves without interpretation from an observer [59]. Examples include assessments of health status and quality of life (e.g. physical ability, symptom severity). PROs are typically recorded in a self-completed questionnaire.

Patient-reported outcome measure

A measure of a PRO including domain (e.g. anxiety), specific measurement tool (e.g. name of questionnaire) and other levels of specification required for clinical outcomes (analysis metric and method of aggregation) [60].

Hospital/process-related outcome

A metric related to the organisation or individual involved in the patients’ care rather than the effect of the intervention on the patient’s health, e.g. length of hospital stay or number of tests conducted.

Resource use measure

A metric to quantify the cost of care, including direct financial expenses as well as staff time.

Surrogate outcome

A measure that is not of direct practical importance but is believed to reflect an outcome that is important; often a physiological or biochemical marker that can be relatively quickly and easily measured, and that is taken as being predictive of an important clinical outcome [3].

Composite outcome

An outcome that consists of two or more component outcomes [5].

Primary outcome

The outcome of greatest importance [3].

Secondary outcome

An outcome used to evaluate additional effects of the intervention deemed a priori as being less important than the primary outcomes [3].

Explanatory outcome

Outcomes that are measured to provide additional information about an intervention, but may be without an a priori hypothesis.